If you are the sort of person who can hold plot lines, multiple sub plots, multiple characters personalities, characters arcs and character developments, in your head all at the same time and still tell a good story, you could write a novel the way you suggest. There are people out there that can do that. You might be one of the few.
The rest us, need to construct a novel in layers, making multiple editing passes, re-structuring, and re-writing until we have knocked the story into shape.
Certainly you can dictate your first draft – I do that a lot, but you need some sophisticated software. I use the Dragon speech to text software from Nuance ( if you buy version 13, rather than 15, you can pick it up for a reasonable price). You can dictate into your mobile and upload it for transcription by the software at a late date. However, to get the most accurate transcription, I find you need a good quality headphone in a quiet environment. I often do it while exercising on an exercise bike or pacing a room. I find, this way, on a first draft, I can produce 3000 words per hour, with 90% accuracy.
I do not know of any speech to text engines that are cloud based, but they are just around the corner. Google, Amazon Polly, and IBM all have cloud based wavenet Artificial intelligences (AI) producing text to speech. Some developers are even experimenting with using these AIs for real time translating, so it is only a matter of time before cloud based speech to text engines appear.
You could use Dragon to verbally edit your drafts too, but I find it easier to use a keyboard, so I switch to a software called Scrivener, because I like to have everything visually laid out before me, but you could use any word processing software.
For my last edit, I reverse the process and use text to speech software, so I can hear the manuscript being ‘read aloud’, paragraph by paragraph. I have been using MS Sienna in MS Word, but I’ve recently been experimenting with producing my own AI read audio books. I have now switched to using Wave Net Vocaliser (not cheap, but a lot cheaper than using an voice artist) so I can hear the read-back in the voice in which it will be narrated. Wavenet is a cloud application.
After the read-back, I will then insert SSML commands into the manuscript for Wave Net Vocaliser, using the Google AI, to convert the whole manuscript into an audio book.
Any or all of these files can be stored and accessed in the cloud.
Hope this helps,